Biased Signaling Favoring Gi over β-Arrestin Promoted by an Apelin Fragment Lacking the C-terminal Phenylalanine*

Background: Apelin receptor represents a therapeutic target for cardiovascular diseases. Results: Apelin 17 activates ERK1/2 in a β-arrestin-dependent and G protein-dependent manner, whereas apelin 17 with deleted C-terminal phenylalanine only signals through the G protein. Conclusion: Biased signaling promoted by an apelin fragment lacking the C-terminal phenylalanine is favoring Gi over β-arrestin. Significance: Apelin receptor β-arrestin signaling may account for apelin hypotensive activity. Apelin plays a prominent role in body fluid and cardiovascular homeostasis. We previously showed that the C-terminal Phe of apelin 17 (K17F) is crucial for triggering apelin receptor internalization and decreasing blood pressure (BP) but is not required for apelin binding or Gi protein coupling. Based on these findings, we hypothesized that the important role of the C-terminal Phe in BP decrease may be as a Gi-independent but β-arrestin-dependent signaling pathway that could involve MAPKs. For this purpose, we have used apelin fragments K17F and K16P (K17F with the C-terminal Phe deleted), which exhibit opposite profiles on apelin receptor internalization and BP. Using BRET-based biosensors, we showed that whereas K17F activates Gi and promotes β-arrestin recruitment to the receptor, K16P had a much reduced ability to promote β-arrestin recruitment while maintaining its Gi activating property, revealing the biased agonist character of K16P. We further show that both β-arrestin recruitment and apelin receptor internalization contribute to the K17F-stimulated ERK1/2 activity, whereas the K16P-promoted ERK1/2 activity is entirely Gi-dependent. In addition to providing new insights on the structural basis underlying the functional selectivity of apelin peptides, our study indicates that the β-arrestin-dependent ERK1/2 activation and not the Gi-dependent signaling may participate in K17F-induced BP decrease.

The apelinergic system is also present in the cardiovascular system (21), and several studies suggest a role for apelin in the control of cardiovascular functions. Apelin increases the contractile force of the myocardium (22)(23)(24), and apelin gene-deficient mice subjected to chronic pressure overload developed severe and progressive heart failure (25). Moreover, intravenous injection of K17F or pE13F in rodents decreases arterial blood pressure (BP) and increases heart rate (7,14,16) via a NO production (26) with a greater effect for K17F than for pE13F (14). In contrast, the deletion of the C-terminal phenylalanine in K17F (K16P) or its substitution by an alanine in pE13F (pE13A) suppresses the ability of these peptides to decrease BP (14,27), demonstrating the central role of the C-terminal Phe for the apelin-induced BP decrease.
We previously showed that the deletion or Ala substitution of the C-terminal Phe in K17F does not modify its ability to bind ApelinR or to inhibit forskolin-induced cAMP production (14,28), indicating that the C-terminal Phe of apelin does not play a major role in apelin binding to the receptor or its capacity to activate G i . In contrast, the ability of K17A and K16P to induce ApelinR internalization is drastically reduced when compared with K17F, suggesting an important role for the C-terminal Phe of apelin in agonist-promoted receptor internalization (14,28).
To understand how the C-terminal Phe of K17F promotes ApelinR internalization, we built a three-dimensional model of the human ApelinR by homology using the validated cholecystokinin-8 receptor type 1 three-dimensional model as a template, and we subsequently docked pE13F or K17F into this model. We identified an hydrophobic cavity at the bottom of the binding pocket in which the C-terminal Phe of pE13F or K17F is embedded by Trp-152 in transmembrane IV and Trp-259 and Phe-255 in transmembrane VI. Site-directed mutagenesis revealed that Phe-255 and Trp-259 are required to trigger ApelinR internalization without playing a role in apelin binding nor in G i protein coupling (28).
The functional dissociation between G i protein coupling and receptor internalization suggests that ApelinR exists in different active conformations depending on the bound ligand leading to the activation of different signaling pathways, which is in line with the concept of biased agonists (29,30). Such agonists stabilize distinct receptor conformations that differ in their signaling partner preference, leading to different biological responses (29,(31)(32)(33).
In the last decade, it has become clear that, in addition to their canonical G protein-and second messenger-dependent pathways, GPCRs also use alternative signaling pathways independently of their coupling to G proteins. One of them involves ␤-arrestins 1 and 2 (also named arrestin 2 and 3), a small family of cytosolic proteins known for their role in GPCR desensitization and internalization (34). Several studies have shown that, in addition to their classical role, ␤-arrestins can also act as signaling scaffolds for many signaling pathways, and in particular those of MAPKs (34), which lead to a second wave of signal transduction (35). Based on these findings, we hypothesized that the important role of the C-terminal Phe of apelin in its hypotensive action may result from a G i -independent but ␤-arrestin-dependent signaling pathway. To test this hypothesis, we took advantage of the apelin fragments K17F and K16P, which exhibit opposite profiles on ApelinR internalization and BP. Using bioluminescence resonance energy transfer (BRET)based biosensors, we first compared the capacity of K17F or K16P to activate G␣ i and promote ␤-arrestin recruitment. We then investigated the ability of K17F and K16P to activate p42/44 MAPK (ERK) in CHO cells stably expressing the rat ApelinR in the absence or presence of pertussis toxin (PTX), a G i protein inhibitor, or in the absence or presence of a mutant of ␤-arrestin (␤-arrestin-2-K296A) that prevents the scaffolding of ERK1/2. Finally, we confirmed the involvement of ␤-arrestin-dependent ERK signaling pathway by using a mutated ApelinR previously shown to be unable to internalize upon K17F stimulation (28).

Animals
All procedures involving animals were carried out in accordance with institutional guidelines for the care and use of laboratory animals. Male Sprague-Dawley rats of 130 -180 g (Charles River Breeding Laboratories, L'Arbresle, France) were used. They were fed a normal standard diet and offered water ad libitum.

Microdissection and Measurement of the Glomerular Arterioles Diameter
Rats were anesthetized by intraperitoneal injection of pentobarbital (60 mg/kg). The left kidney was prepared for microdissection of arterioles as previously described (36). Afferent arterioles were microdissected attached to the glomeruli. Sequential photographies were recorded on the same arteriole with a digital camera (Coolpix 5400; Nikon) under three experimental conditions at 1.5-min intervals: basal condition, 10 Ϫ9 M Ang II and 10 Ϫ9 M Ang II ϩ 10 Ϫ7 M K17F or ϩ 10 Ϫ7 M K16P. Diameters were measured on a distance equal to 20, 60, and 100 m upstream of the glomerulus with ImageJ 1.43u, and the average diameter was calculated. Calibration was made using a stage micrometer.

Constructs
The wild-type and mutated (F255A) rat ApelinRs tagged at their C-terminal part with YFP were constructed by subcloning the open reading frame of the wild-type or F255A ApelinR from previous constructs encoding the wild-type or F255A rat ApelinR tagged at its C-terminal part with EGFP (ApelinR-EGFP) (4,28) to pYFP-N1 (Clontech) using HindIII-BamHI restriction sites. C terminus HA-tagged wild-type and F255A-ApelinR were made by PCR using as template wild-type and F255/A-ApelinR-EGFP constructs and 5Ј-CATTGACGCAA-ATGGGCGGTAGGCGTG-3Ј and 5Ј-GGATCCTAAGCGT-AATCGGGCACATCGTAAGGGTAAGCGTAATCGGGC-ACATCGTAAGGGTAGTCCACAAGGGTTTCTTGGCTA-TAG-3Ј as forward primer and reverse primer, respectively. The PCR products were then digested by HindIII-BamHI restriction enzymes (New England Biolabs) and cloned into pcDNA3 (Invitrogen). Wild-type rat ␤-arrestin 2 fused at its C terminus with DsRed was constructed by amplifying rat ␤-ar-restin2 cDNA by PCR using 5Ј-CTAGCTAGCACCATGGGT-GAAAAACCCGGGAC-3Ј and 5Ј-CCCAAGCTTGCAGAAC-TGGTCATCACAGTC-3Ј as forward and reverse primers, respectively, and by subcloning the PCR product into pDsRedmonomer-N1 (Clontech) using NheI-HindIII restriction sites. K296/A rat ␤-arrestin 2-DsRed was made by performing a first PCR to generate a megaprimer using 5Ј-TGGGCAACTCGCG-CACGAAGACA-3Ј (underlined nucleotides encode the alanine residue replacing the natural lysine residue) and 5Ј-CCC-AAGCTTGCAGAACTGGTCATCACAGTC-3Ј as forward and reverse primers, respectively. A second PCR was then performed using the product of the first PCR as a megaprimer (reverse primer) and the 5Ј-CTAGCTAGCACCATGGGTGA-AAAACCCGGGAC-3Ј as a forward primer. The sequences of all constructs were checked by sequencing.
The G␣ i -RlucII was previously described by Breton et al. (37). G␤ 1 and G␥ 5 coding pcDNA3.1Ϯ vectors were obtained from Missouri University of Science and Technology. A GFP10-G␥ 5 coding vector was subsequently produced by inserting the GFP10 sequence using the restriction sites NheI and Acc65I. The EPAC-based cAMP biosensor human ␤-arrestin1-RlucII and ␤-arrestin 2-RlucII plasmids were described by Breton et al. (37) and Zimmerman et al. (38).

G␣ i Activation Assay
To directly monitor G␣ i activation by ApelinR, we monitored the BRET between the RlucII-tagged G␣ i and the GFP10tagged G␥ 5 subunits, which are known to separate after receptor's activation (39). Briefly, HEK293T cells were transfected with plasmids coding for G␣ i1 -RlucII, G␤ 1 , and G␥ 5 -GFP10 along with the wild-type HA-tagged ApelinR constructs. 24 h post-transfection, the cells were detached in PBS and seeded into a white opaque 96-well plate (PerkinElmer Life Sciences) at a density of 8 to 10 ϫ 10 4 cells/well. 48 h post-transfection, varying concentrations of K16P or K17F were added for the indicated times to the wells along with the luciferase substrate DeepBlue C (2.5 mM final concentration; Goldbiotechnology, Inc., St. Louis, MO). The plates were then read in a Mithras LB940 instrument (Berthold Technologies, Bad Wildbad, Germany), and data were collected using the MicroWin 2000 software (Berthold Technologies). BRET signals were determined by calculating the ratio of the light emitted at 495-535 nm (GFP10) to the light emitted at 330 -470 nm (Luciferase). This BRET signal is defined as BRET2.
cAMP Production Measurements cAMP production was monitored using the EPAC-BRET biosensor as previously described (37). Briefly, HEK293T cells were transfected with the EPAC-BRET construct along with the wild-type ApelinR coding plasmid. 24 h post-transfection, the cells were seeded at a density of 8 -10 ϫ 10 4 cells/well in white opaque 96-well plates in PBS. 48 h post-transfection, vehicle, forskolin (10 Ϫ5 M) or forskolin (10 Ϫ5 M) ϩ K17F or K16P (10 Ϫ6 M) were added to the cells, and the plates were incubated at 37°C for 15 min prior to BRET reading. Luciferase substrate DeepBlue C (2.5 ϫ 10 Ϫ6 M final, Goldbiotechnology, Inc.) was added 5 min prior to reading. BRET2 signals were determined by calculating the ratio of the light emitted at 495-535 nm (GFP10) to the light emitted at 330 -470 nm (Luciferase) using the Mithras LB940 apparatus. BRET2 values obtained from vehicle-treated wells were subtracted from values obtained in the presence of forskolin and forskolin ϩ ligands to yield ligand-induced ⌬BRET2 values that represent ligand promoted cAMP production. A decrease in BRET2 signal represents an increase in cAMP production.

␤-Arrestin Recruitment Assays
The interactions between ␤-arrestins and ApelinR were monitored by BRET, as previously described (40). Briefly, the ␤-arrestin1-RlucII or the ␤-arrestin2-RlucII-encoding plasmids were transfected along with the YFP-tagged wild-type ApelinR construct in HEK293T cells. 24 h post-transfection, the cells were transferred into a white opaque 96-well plate at a density of 8 -10 ϫ 10 4 cells/well. At 48 h post-transfection, the medium was replaced with PBS, and the cells were exposed to varying concentration of either K17F or K16P for 15 min prior to BRET reading to yield dose-response curves. The RLuc substrate coelentherazine H (2.5 M; Nanolight Technology, Pinetop, AZ) was added 5 min prior to reading. For kinetic studies, Coelentherazine H was first added. Five min following the substrate addition, wells were treated with either vehicle, K16P-or K17F-apelin (both at 10 Ϫ6 M), and the plates were immediately read repetitively for 15 min in the Mithras LB940 instrument. The BRET signal is determined by calculating the ratio of the light emitted at 505-555 nm (YFP) to the light emitted at 465-505 nm (Luciferase). This signal is defined as BRET1.

ERK1/2 Phosphorylation
Assessed by Western Blotting-CHO-K1 cells stably expressing wild-type ApelinR were grown in 12-well plates and starved for 16 h in a serum-free medium without or with PTX (25 ng/ml) prior to different stimulations: 10 Ϫ7 M K17F or K16P for time course experiments, different concentrations from 10 Ϫ11 to 10 Ϫ6 M of K17F or K16P for 10 min for dose-response experiments. After stimulation, the medium was removed, 100 l of SDS sample buffer (pH 6.8) containing 10 mM of dithiothreitol was added to each well, and the cells were placed on ice. Whole cell lysates were sonicated, heated for 5 min at 95°C, then resolved on 10% Tris/glycine polyacrylamide gel, and transferred onto polyvinylidene difluoride membranes for immunoblotting. The membranes were incubated overnight with a 1:1000 dilution of phospho-ERK1/2 antibodies in 3% (w/v) bovine serum albumin Tris-buffered saline Tween (TBST) at 4°C. The membranes were subsequently washed three times in TBST for 10 min and then incubated with a 1:10000 dilution of HRP-conjugated antibodies for chemiluminescence detection (Amersham Biosciences) for 1 h at room temperature. For the total ERK1/2 assessment, the membranes were stripped and reprobed using anti-mouse ERK1/2 antibodies (dilution 1:1000). ERK phosphorylation was normalized according to the loading of proteins by expressing the data as a ratio of pERK1/2 over total ERK1/2. Chemiluminescence detection was performed using ECL (GE Healthcare) and was quantified by densitometry with the ImageJ gel analysis software (Image Processing and Analysis in Java). ERK1/2 Assessed by Alphascreen-CHO-K1 cells stably expressing the wild-type or F255A mutated ApelinR-EGFP were plated in 96-well plates at a density of 3 ϫ 10 4 cells/well and grown for 24 h. The cells were then starved for 16 h in a serum-free medium with or without PTX (25 ng/ml) and treated with K17F or K16P at 10 Ϫ7 M for different times. ERK1/2 phosphorylation was measured using the ERK1/2 Alphascreen Surefire kit (PerkinElmer Life Sciences). After stimulation by K17F or K16P, cells were lysed on ice in the lysis buffer provided by the manufacturer. 4 l of lysates were then transferred to a 384-well white Optiplate (PerkinElmer Life Sciences), and ERK1/2 activation was measured according to the manufacturer's instructions. ERK1/2 phosphorylation values obtained from wells treated with vehicle were subtracted from values obtained in presence of ligands to get specific ligandpromoted ERK phosphorylation values.

Bias Calculation
All of the data were analyzed using the nonlinear curve fitting equations in GraphPad Prism (v6.0) to estimate the pEC 50 values of the curves for the different pathways. Ligand bias was quantified by analyzing the concentration-response curves using the operational model of agonism, as described previously (41)(42)(43).
The operational model was also used to determine the transduction ratios (/K A ) of both K17F and K16P as previously described (42)(43)(44)(45) using the following equation, where E is the effect of the ligand, [A] is the concentration of agonist, E m is the maximal response of the system, Basal is the basal level of response in the absence of agonist, LogK A is the logarithm of the functional equilibrium dissociation constant of the agonist, n is the slope of the transducer function that links occupancy to response, and LogR is the logarithm of the "transduction coefficient" (or "transduction ratio"), /K A , where is an index of the coupling efficiency (or efficacy) of the agonist.
To eliminate the influence of the system and the observation biases, the activity of K16P at a given signaling pathway was compared with that of K17F used as the reference compound using the following equation.
Ligand bias was calculated using the following, where L1 and L2 are ligands 1 and 2, respectively; P1 is pathway 1; and P2 is pathway 2 as recently described (45). The bias between two ligands was expressed as 10 ⌬⌬log( /K A ).

Statistical Analysis
Statistical significance of the difference between conditions for BRET assays were assessed by performing two-way analysis of variance followed by Bonferroni post-tests using Prism 5.0 software (GraphPad Software, La Jolla, CA). Statistical comparisons for ERK1/2 phosphorylation experiment data were analyzed with GraphPad Prism and were performed using the paired Student's t test.
For the pairwise comparisons between K16P and K17F signaling bias, statistical analysis was performed using a two-way unpaired Student's t test on the ⌬Log(/KA). Differences were considered significant when p was Ͻ 0.05.

Effects of K17F and K16P on Wild-type Apelin
Receptor-mediated G␣ i Activation-To characterize the effects of K17F and K16P on ApelinR-mediated G␣ i signaling, HEK293 cells were transiently transfected with the HA-tagged wild-type rat ApelinR. BRET2-based assays were used to assess the activation of the G i protein by monitoring the dissociation of the G␣ i -G␥ 5 complex and to investigate the inhibition of forskolin-induced cAMP production using the BRET2-based EPAC biosensor (46). Dose-response curves of K17F-and K16P-promoted G i activation (Fig. 1A) showed very similar activities of both peptides to activate G i as illustrated by the ability of the peptides to inhibit the G␣ i1 -G␥ 5 BRET2 signal with comparable maximal responses and similar efficacies (IC 50 values: 2.7 Ϯ 0.52 10 Ϫ10 and 8.1 Ϯ 0.52 10 Ϫ10 M, respectively). The similar efficacy of the two peptides for G i activation was further confirmed by the similar ability of K17F and K16P to inhibit forskolin-stimulated cAMP production, as illustrated by the blockade of the forskolin-promoted decrease in EPAC BRET2 signal (Fig. 1B).
Effects of K17F and K16P on the Recruitment of ␤-Arrestin 1 or ␤-Arrestin 2 by the Wild-type Apelin Receptor-Recruitment of ␤-arrestin 1 and 2 by ApelinR following K17F or K16P treatment was then investigated. To directly assess the physical interaction between the ligand-bound ApelinR and ␤-arrestin 1 or 2, the wild-type rat ApelinR fused to YFP at its C terminus and the human ␤-arrestin 1 or ␤-arrestin 2 fused to RLucII were transiently coexpressed in HEK293 cells. Fig. 2 (A and B, upper  panels) shows that K17F and K16P increased the BRET1 signal for both ␤-arrestins 1 and 2 recruitment in a dose-dependent manner. Interestingly a greater recruitment was observed for ␤-arrestin 2 versus ␤-arrestin 1 This difference in BRET1 signal is consistent with the previously documented reduced ability of class A GPCRs to recruit ␤-arrestin 1 versus ␤-arrestin 2 (47). However, the lower signal could also represent different conformations for the ␤-arrestin/receptor complex for the two ␤-arrestin isoforms. For both ␤-arrestin 1 and ␤-arrestin 2, the maximal response was significantly higher for K17F than for that observed with K16P. Despite the significant difference in the efficacy of the two apelin peptides to promote ␤-arrestin recruitment, they showed similar potencies (EC 50    mote the recruitment of ␤-arrestins is lower than that observed for the activation of G i and cAMP production. This most likely reflects the different levels of amplification of the assays used but could also reflect a preference of the ligand-bound receptor for the G i pathway. Kinetic experiments for both ␤-arrestins 1 and 2 recruitment reported in Fig. 2 (A and B, lower panels) confirmed the significantly greater propensity of K17F to promote ␤-arrestins recruitment compared with K16P. Indeed, K17F induced a rapid and strong increase in the BRET1 signal, whereas K16P induced a slower and much smaller increase for both ␤-arrestins. To compare the relative efficiency of K16P and K17F to engage ␤-arrestin 1 and 2 versus G i , biased signaling was quantified using the operational model (see "Experimental Procedures" for details). As shown in Table 1, K16P was found to be 8.9-and 6.1-fold less efficient to engage ␤-arrestin 2 and 1 than K17F, whereas the two peptides had similar ability to activate G i .
Effects of K17F and K16P on the Wild-type Apelin Receptormediated ERK1/2 Activation-To compare the ability of K17F and K16P to induce ERK1/2 phosphorylation, CHO-K1 cells stably expressing the wild-type ApelinR-EGFP were treated with either K17F or K16P (10 Ϫ7 M) for different times (from 5 to 120 min) (Fig. 3, A and B) or treated with increasing concentrations of K17F and K16P (from 10 Ϫ11 to 10 Ϫ6 M) for 10 min (Fig.  3, C and D). Cells were then lysed, and ERK1/2 phosphorylation was measured by Western blot analysis using anti-phospho-ERK antibodies. As shown in Fig. 3A, K17F treatment led to a significant ERK1/2 phosphorylation. A similar ERK1/2 phosphorylation profile was observed following K16P treatment. K17F-or K16P-induced ERK1/2 phosphorylation reached a peak value at 5-10 min and remained greater than 2.5-fold over basal levels for at least 60 min. Dose-response curves of ERK1/2 phosphorylation in response to K17F or K16P treatment (Fig. 3,  C and D) showed a similar maximal effect with an EC 50 value for K17F of 5.6 Ϯ 1.8 ϫ 10 Ϫ11 M, 13-fold lower than that of K16P (7.4 Ϯ 2.6 ϫ 10 Ϫ10 M).

Involvement of a G i Protein-dependent and -independent Mechanism in ERK1/2 Phosphorylation Induced by K17F and
K16P-To determine the contribution of a G␣ i -dependent mechanism to trigger K17F-induced ERK1/2 phosphorylation, we assessed the effect of a PTX treatment (25 ng/ml) for 16 h prior to stimulation of apelinR-expressing CHO-K1 cells with K17F (10 Ϫ7 M). Time course analysis of ERK1/2 phosphorylation induced by K17F in absence of PTX showed a peak of maximal phosphorylation at 5-10 min and then a progressive decrease from 30 to 120 min. In contrast, in the presence of PTX, the ERK1/2 phosphorylation induced by K17F was strongly decreased but remained present up to 25 min post-stimulation (Fig. 4, A and B). The effect of the PTX treatment was more drastic on the K16P-induced ERK1/2 phosphorylation. Indeed, although some residual activity could be detected

/K A ) ratios and bias factors for K17F and K16P
Statistical analysis was performed using a two-way unpaired Student's t test on the ⌬log(/K A ).  A, CHO cells stably expressing wild-type apelin receptor were treated with 10 Ϫ7 M of K17F or K16P for 120 min. ERK1/2 phosphorylation was detected by Western blotting analysis using anti-phospho-ERK1/2 antibody (p-ERK), and total ERK1/2 was detected by total anti-ERK1/2 antibody (total ERK) in the same immunoblot for loading control. B, quantification of the bands corresponding to 44 and 42 kDa was performed by densitometry using ImageJ software. The data are expressed as percentages of pERK/total ERK. C and D, dose response of apelin receptormediated ERK1/2 phosphorylation induced by K17F and K16P stimulation. C, cells were treated with the indicated concentrations of K17F and K16P for 10 min, and ERK1/2 phosphorylation was measured. D, quantification of Western blotting bands was performed by densitometry analysis. The data correspond to the means Ϯ S.E. from at least three independent experiments. at 10 min, it was totally abolished after 20 min (Fig. 4C). The different effect of PTX treatment on K17F and K16P responses is particularly well illustrated in Fig. 4 (E and F), respectively, in which ERK1/2 phosphorylation induced by K17F (10 Ϫ7 M) in the presence of PTX was monitored by the more quantitative Alphascreen assay. As shown in Fig. 4 (E and F), the level of ERK1/2 phosphorylation in PTX-treated cells measured over a period of 10 min was 3.4 times higher than vehicle upon stimulation with K17F (Fig. 4E), whereas only a marginal Ͻ2-fold increase, which did not reach statistical significance, was observed upon stimulation with K16P (Fig. 4F). Interestingly, the PTX-resistant response corresponds to the early phase of the K17F-promoted ERK1/2 response, suggesting that the latter phase is entirely G i -dependent. This is in contrast with what was observed for other GPCRs such as the AT1 receptor, for which ␤-arrestin 2 has been shown to contribute to the late phase of the ERK1/2 response (34,48,49). However, a lack of a positive contribution of ␤-arrestin to late ERK1/2 response has been previously reported for the M3-muscarinic receptor (50).

Involvement of ␤-Arrestin 2 Recruitment in K17F-induced G␣ i Protein-independent ERK1/2 Phosphorylation-To investigate whether ERK1/2 phosphorylation could in part involve
␤-arrestin 2 recruitment to the receptor, CHO-K1 cells stably expressing the wild-type ApelinR-EGFP were transiently transfected with a control empty vector (DsRed mono), or a mutated form of ␤-arrestin 2 that inhibits ␤-arrestin-2 self-association, prevents the scaffolding of ERK1/2 and cannot support the ␤2-adrenergic receptor-stimulated ERK1/2 (DsRed mono-␤arrestin 2-K296A) (51). Cells were thus stimulated for 5-20 min with K17F or K16P (10 Ϫ6 M) following pretreatment with PTX (25 ng/ml) or vehicle, and ERK1/2 phosphorylation was quantified using Alphascreen technology (Fig. 5, A-D). Under basal condition, ERK1/2 phosphorylation induced by K17F or K16P is not significantly affected by the transfection of either DsRed mono or ␤-arrestin 2 K296A. As expected, treatment with PTX reduced ERK1/2 phosphorylation induced by K17F (a factor of 4.2) or K16P (a factor of 9.6) (Fig. 5, A and C), yielding responses that remained significantly different from vehicle for K17F but not for K16P, which further illustrates the greater importance of the G i component for the K16P-stimulated ERK pathway. Interestingly, in cells treated with PTX, expression of the ␤-arrestin 2 K296A completely abolished the K17F-induced ERK1/2 phosphorylation (Fig. 5, B and D), highlighting the significant ␤-arrestin-dependent component of the K17F-stimulated ERK1/2 response.
Effects of K17F and K16P on ERK1/2 Phosphorylation in Cells Expressing the F255A Mutated Apelin receptor-We then investigated the capacity of the F255A mutated ApelinR (a mutant form of the receptor that does not undergo internalization upon K17F stimulation) to trigger ERK1/2 phosphorylation. We compared the ability of K17F and K16P to induce ERK1/2 phosphorylation following stimulation of the wild-type or the F255A mutated ApelinR in absence or in presence of PTX (Fig. 6, A-D). In the absence of PTX treatment, K17F and K16P induced a strong and similar ERK1/2 phosphorylation with a 26-fold increase over vehicle in CHO-K1 cells stably expressing the wild-type ApelinR (Fig. 6, A and C), whereas a significantly blunted ERK1/2 phosphorylation was observed in cells stably expressing the F255A mutated ApelinR, with a 6.9and 3.6-fold increase for K17F and K16P, respectively, over vehicle (Fig. 6, B and D). As reported above, pretreatment with PTX significantly inhibited both K17F-and K16P-induced ERK1/2 phosphorylation in cells expressing the wild-type ApelinR (Fig. 6, A and C). Whereas the K16P-stimulated response was fully inhibited, a 4-fold increase over vehicle was still observed for K17F. This PTX-resistant response was fully abolished in cells expressing the F255A mutated ApelinR. Indeed, neither K17F nor K16P promoted ERK1/2 phosphorylation in PTX-treated cells expressing F255A-ApelinR (Fig. 6, B  and D).
Vasoactive Role of K17F versus K16P in Rat Glomerular Afferent Arterioles-We finally investigated the ability of K17F and K16P to induce vasorelaxation of glomerular afferent arterioles preconstricted by Ang II (Fig. 7). Treatment of isolated glomerular afferent arterioles with 10 Ϫ9 M of Ang II induced a vasoconstriction of the arterioles because we observed a significantly reduced arteriole diameter (12.

DISCUSSION
In the present work, we show that the C-terminal Phe of apelin plays a critical role in ␤-arrestin recruitment to the ApelinR but is not involved in G␣ i protein activation nor adenylyl cyclase inhibition. We also demonstrate that apelin-induced ERK1/2 MAPK activation involves both G␣ i -and ␤-arrestindependent pathways and that the ␤-arrestin-dependent ERK1/2 phosphorylation requires the presence of the C-terminal Phe of apelin. We also show that stimulation by K17F of an ApelinR in which the residue Phe-255 (predicted to interact with the C-terminal Phe of K17F (28)) is replaced by an alanine, did not internalize, and did not induce G protein-independent ERK1/2 phosphorylation. Together, these data demonstrate that stimulation of ApelinR by K17F but not by K16P activates a G␣ i protein-independent but ␤-arrestin-dependent signaling pathway leading to MAPK activation. Previous pharmacological studies showed that the C-terminal Phe in K17F does not play a major role in determining apelin binding affinity or the ability of the apelin peptides to inhibit forskolin-induced cAMP production (11,14,28) but is crucial to trigger ApelinR internalization. In addition, the deletion of the C-terminal Phe in K17F (K16P) or its substitution by an Ala in pE13F (pE13A) suppresses the ability of these peptides to decrease BP (14,27), demonstrating the central role of the C-terminal Phe for the apelin-induced BP decrease. Altogether, this suggests that the decrease in BP requiring the C-terminal Phe of apelin may result from a G i -independent but ␤-arrestindependent signaling pathway. To test this hypothesis, we first characterized the capacity of K17F and K16P to activate G␣ i and to induce ␤-arrestin 1/2 recruitment. We showed that K17F and K16P are potent inducers of G␣ i protein activation and adenylyl cyclase inhibition, in agreement with their effects on the inhibition of forskolin-induced cAMP production (28). In contrast, whereas K17F was found to be a potent inducer of ␤-arrestin 1/2 recruitment, K16P had a very low activity toward this pathway, even at a supramaximal concentration of 10 Ϫ6 M. These data further support that K16P behaves as a biased ligand favoring G protein-coupling versus ␤-arrestin recruitment, consistent with the previous study demonstrating that K16P does not promote ApelinR internalization (28). This bias is clearly evident with the ␤-arrestin 2 recruitment-based assay as previously observed with the internalization-based assay. Indeed, the maximal BRET signal obtained with K16P at a supramaximal concentration of 10 Ϫ6 M is decreased by 60% as compared with that obtained with K17F at the same concentration. Formal quantification of the bias revealed a 8.9-fold lower efficiency of K16P to promote ␤-arrestin 2 engagement than K17F, whereas the two peptides have equivalent abilities to activate G␣ i . Taking into account the respective effects of these two peptides on BP, it is therefore tempting to propose that the apelin-induced BP decrease promoted by K17F results from the activation of one or several G␣ i -independent but ␤-arrestin-dependent signaling pathway(s).
Because ␤-arrestin acts a scaffold protein that plays a role in the activation of MAPK, we hypothesized that the stimulation of ApelinR by K17F, inducing ␤-arrestin recruitment, should result in the activation of MAPK as previously shown for the  angiotensin II receptor type 1a (AT1a) (49). We showed here that K17F is 13 times more potent than K16P to activate ERK1/2 phosphorylation, in agreement with the reduced ability of the peptide to promote ␤-arrestins engagement and ApelinR internalization (28). Altogether, these findings suggest that the difference in efficiency of both peptides to activate ERK1/2 phosphorylation could be linked to their differential ability to induce ␤-arrestin 1 or 2 recruitment and receptor internalization. This hypothesis is supported by our observation that although PTX-promoted inactivation of G␣ i led to a strong decrease in both K17F-and K16P-stimulated ERK1/2 phosphorylation. However, there was a significant PTX-resistant response evoked by K17F, even if the total of ERK1/2 phosphorylation is mainly G i -dependent. This PTX-resistant response occurs in a short window of time between 10 and 20 min after stimulation and disappears in the long-term sustained ERK1/2. This was in contrast with the complete abrogation by PTX treatment of the pE13F-induced ERK phosphorylation (13,52), which is in line with the observed lower efficiency of pE13F as compared with K17F to induce ApelinR internalization and to decrease BP (14). In addition, the PTX-resistant ERK1/2 phosphorylation evoked by K17F was completely blocked by ␤-arrestin-2-K296A, which does not bind ERK1/2 efficiently (51). The observation that the F255A mutated ApelinR, which does not undergo endocytosis upon K17F stimulation while keeping its capacity to activate G␣ i coupling (28), was unable to trigger ERK1/2 phosphorylation further confirms that the G protein-independent ERK1/2 phosphorylation is mediated by ␤-arrestin recruitment.
Altogether, we demonstrated that ApelinR MAPK signaling occurs through at least two distinct signaling pathways: one dependent on G␣ i activation and the other dependent on ␤-arrestin recruitment. This dual signaling property of GPCRs has been reported for several receptors (35,48,(53)(54)(55)(56)(57)(58) and has given birth to the concept of biased agonists. This implies that a specific ligand might trigger distinct receptor conformations leading to different signaling pathways through interaction with different effector proteins. Such a mechanism has been shown for ␤2-adrenergic receptor (59) and more recently for V2 receptor (60). In this context, K16P could be considered as a G protein-biased agonist that conserves its ability to activate G protein signaling, whereas it lacks its capacity to trigger ␤-arrestin signaling. K17F and K16P will thus trigger different signalings by inducing different ApelinR conformations.
The physiological consequences of a G protein-independent and ␤-arrestin-dependent signaling pathway has been illustrated for AT1a receptor in cardiomyocytes. Stimulation by Ang II or [Sar 1 , Ile 4 , Ile 8 ] Ang II of AT1a receptor led to positive inotropic and lusitropic responses in wild-type cardiomyocytes, whereas this response was abolished in ␤-arrestin 2 knock-out cardiomyocytes (61). The therapeutic potential of such compounds has been illustrated by the development of TR120027, a AT1a receptor ␤-arrestin-biased agonist, which improved cardiac performance and had cardioprotective activity by blocking the G protein-mediated vasoconstriction while increasing the cardiac contractility through ␤-arrestin signaling (62,63). The activation of ApelinR ␤-arrestin-mediated signaling could have also a biological role because only apelin frag-ments that fully activate ␤-arrestin induced a decrease in arterial BP (14). So far, the lack of ␤-arrestin recruitment following K16P application has never been demonstrated in cells that endogenously express apelin receptors. Nevertheless, we recently observed in glomerular afferent arterioles from rat kidney, expressing apelin receptor mRNA (36) that 10 Ϫ7 M K17F significantly (p Ͻ 0.001) induced a vasodilatation of afferent arterioles preconstricted by 10 Ϫ9 M angiotensin II. In contrast, K16P, which is biased against ␤-arrestin-dependent signaling, is ineffective, suggesting a direct link between the biased signaling and the differential vasodilatory response promoted by distinct apelin fragments. In addition, Scimia et al. (64) have recently shown that a mechanical stretch of cardiomyocytes activates ERK1/2 phosphorylation in an ApelinR-dependent manner, independently of G␣ i activation, and promotes ␤-arrestin recruitment. ERK1/2 activation by ␤-arrestin-dependent signaling pathway leads to a redistribution of ERK1/2 into endosomal vesicles that contain receptor-␤-arrestin complexes as previously shown for AT1a receptor (65). Such redistribution of ERK1/2 might occur in endothelial cells or in cardiomyocytes upon ApelinR activation and might contribute to the actions of apelin on blood vessels and heart.
In conclusion, we have demonstrated that ApelinR triggers ERK1/2 phosphorylation in both a G protein-and ␤-arrestindependent manner. We further showed that K16P is a G protein-biased agonist displaying a strongly impaired ␤-arrestin signaling pathway that may account for its lack of activity in vivo on arterial blood pressure. This study brings new insights into ApelinR signaling mechanisms. Further studies are required to better understand the ApelinR molecular signaling features required for the development of ApelinR biased agonists that could be therapeutically useful. The development of such compounds will be interesting to further explore the physiopathological roles of apelin and to be able to target a specific apelin biological action.